EnEfG and AI Power: What Data Centers Must Consider Now

8 min Read Time

Germany’s data centers now consume over 21 billion kilowatt-hours of electricity annually – the first time this threshold has been crossed. AI workloads are accelerating demand further: installed GPU capacity is projected to quadruple by 2030. At the same time, the Energy Efficiency Act (EnEfG) tightens regulatory requirements: from July 2026, new data centers must comply with a PUE limit of 1.2. For Germany’s SMEs – increasingly reliant on cloud and AI infrastructure – energy efficiency has become a decisive factor in location decisions.

The Key Takeaways

What the EnEfG Requires from Data Centers

The Energy Efficiency Act (EnEfG), effective since November 2023, is Germany’s first law to impose binding efficiency standards for data centers. It applies to all facilities with a non-redundant rated connection capacity exceeding 300 kilowatts – including most commercial data centers and larger in-house server rooms operated by enterprises.

Requirements roll out in three phases:

From July 2026: New data centers must demonstrate a PUE of no more than 1.2. PUE (Power Usage Effectiveness) measures the ratio of total facility power consumption to IT equipment power consumption. A PUE of 1.2 means only 0.2 kWh of ancillary power – for cooling, UPS systems, and infrastructure – is required per 1 kWh of IT load. That target is ambitious: according to the Borderstep Institute, Germany’s current average PUE stands at approximately 1.55.

From July 2027: Existing data centers must meet a PUE of 1.5. For legacy installations relying on air cooling and inefficient UPS systems, this will necessitate infrastructure upgrades.

From July 2030: Existing data centers must improve to PUE 1.3. This second phase grants operators time for major retrofits while establishing a clear transformation roadmap.

AI Power: Why GPU Workloads Are Changing the Equation

The energy dynamics within data centers are shifting fundamentally. Traditional CPU workloads consume 65-250 watts per processor. A single NVIDIA H100 GPU draws up to 700 watts at full load. A typical AI training node with eight GPUs consumes 10-12 kilowatts. A single rack housing eight to ten such nodes reaches 80-140 kW – five to ten times the load of a conventional server rack.

The consequence: AI workloads are overwhelming existing cooling concepts. Air cooling hits physical limits at rack power densities above 30 kW. Liquid cooling is no longer optional – it’s essential.

According to Bitkom, Germany currently hosts 530 megawatts of AI-capable data center capacity – about 15% of total national capacity. By 2030, that share is expected to grow to 2,020 megawatts. Goldman Sachs forecasts a global 165% increase in data center electricity consumption by 2030, driven primarily by GPU-intensive AI applications.

At the same time, GPUs deliver significantly higher energy efficiency than CPUs for AI tasks: the same computation requiring hundreds of times more energy on CPUs can be completed by a GPU in a fraction of the time. Thus, rising overall consumption stems not from inefficiency – but from the sheer volume of new applications.

“Data centers could, in the long term, supply over 20% of Frankfurt’s municipal heating demand. The waste heat generated would theoretically suffice to heat every residential building in the city.”
Paraphrased from the Borderstep Institute, Waste Heat Utilization in Data Centers, 2022/2024

Waste Heat Reuse: From Cost Factor to Business Model

The EnEfG introduces, for the first time, a mandatory waste heat reuse requirement for data centers. Requirements escalate in stages: an Energy Reuse Factor (ERF) of 10% from July 2026, 15% from 2027, and 20% from 2028. Acceptable forms include internal reuse, delivery to third parties, and feed-in to municipal district heating networks.

In practice, this creates a new business model: data centers as neighborhood heat suppliers. In Frankfurt – the largest German data center hub – waste heat could eventually cover more than 20% of the city’s heating demand. According to the Borderstep Institute, over 13 billion kilowatt-hours of data center waste heat are released unutilized into the environment across Germany each year.

For operators running their own data centers or server rooms, waste heat reuse offers a dual benefit: it fulfills statutory obligations while simultaneously reducing heating costs for adjacent office spaces or production halls.

What Retrofitting Costs

Investment needs depend heavily on the starting point. For typical SME data centers, three scenarios are most relevant:

Air cooling optimization (€20,000-€80,000): Cold aisle containment, raising supply air temperature to 27°C, and airflow optimization. Often sufficient to improve PUE from 1.6-1.8 to 1.5.

Indirect free cooling (€100,000-€300,000): Leveraging ambient outdoor air for cooling when temperatures permit – in Germany, typically six to nine months per year. Lowers PUE to 1.2-1.3 and cuts cooling-related electricity consumption by 40-60%.

Liquid cooling for AI racks (€250,000-€800,000 per installation): Direct-to-chip or immersion cooling for high-density GPU racks. According to Datacenter-Insider, additional investment versus air cooling totals around €739,000, with annual savings of €380,000 – achieving payback in under two years.

In Practice: How Equinix Frankfurt Lowers PUE

Equinix operates several data centers in Frankfurt and demonstrates how EnEfG requirements are implemented in practice. At the FR6 data center, energy consumption was reduced by up to 9% through the use of AI-based cooling control from Darmstadt-based startup etalytics.

The technology analyzes weather data, utilization patterns, and cooling loops in real-time, automatically optimizing cooling performance. The result: PUE values between 1.2 and 1.35 – compliant with EnEfG requirements for new builds. Equinix plans to roll out the technology to additional locations.

Pilot projects are also emerging in waste heat utilization: ENGIE Refrigeration cooperates with municipal utilities at over ten German data centers to feed waste heat into local district heating networks. Around 14-15% of German households are connected to district heating networks – the infrastructure for waste heat utilization is already in place in many locations.

What SMEs Should Do Now

1. Take stock. Measure PUE (don’t estimate), document cooling systems, record power consumption per rack. You can’t plan if you don’t know your current PUE.

2. Check registration. Data centers over 300 kW must be registered in the BfEE Energy Efficiency Register since July 2025. Annual data submission is due by March 31.

3. Implement an energy management system. Mandatory since January 2026. ISO 50001 or EMAS as a framework. Many companies can dock this onto an existing governance system.

4. Evaluate cooling strategy. PUE over 1.5? Evaluate cold aisle containment and free cooling. Planning AI workloads? Plan liquid cooling early. The payback is economically attractive given rising electricity prices and growing power density.

5. Check waste heat potential. Contact municipal utilities or neighboring businesses. The waste heat obligation from 2026 (10% ERF) can often be met with simple heat exchanger solutions for adjacent buildings.

Conclusion

The EnEfG makes energy efficiency in data centers mandatory rather than optional. AI-driven demand for GPU capacity further intensifies the challenge. For SMEs, this means: If you operate your own server rooms or colocation spaces, you must act by July 2027. The good news: Investments in more efficient cooling pay off quickly, and waste heat utilization even opens up new revenue streams. Energy efficiency in the data center is no longer an environmental topic in 2026 – it’s an economic issue.

Frequently Asked Questions

Header Image Source: Pexels / Manuel Geissinger